Learn · Enforcement
Oathera's own rules: observe first, then enforce
A new rule shouldn't break a working agent on its first day. Oathera's rules start by watching: you see what each one would refuse before it refuses anything.
What a rule decides
Every call an agent makes through Oathera Agent Enforcement is checked first: the identity token, the operation proof, the agent's fingerprint. Then your rules answer: allow, deny or approval required. A rule matches on what the call does (its capability, such as read or write), the operation and the tool, the audience, whether the person behind the agent is still active, and facts about the agent's host. “This agent may read the sales data, not write it” is one rule.
Observe first, then enforce
Each rule has a mode: off, observe or enforce, and observe is the default. An observing rule never changes the answer. It logs what it would have done, so you see which calls it would refuse before it refuses any. When the log shows the rule is right, switch it to enforce.
When rules disagree
Among enforced rules, deny beats approval required, and approval required beats allow.
Where the rules run
In Oathera Agent Enforcement on your own MCP server, after the identity checks and before the call reaches your tool. Your identity provider and your agent directory stay as they are.
FAQ
Does a new rule refuse anything on day one?
No. A rule starts in observe: it logs the calls it would refuse and changes nothing until you switch it to enforce.
What can a rule match on?
The call's capability, such as read or write, the operation, the tool, the audience, whether the person behind the agent is still active, and facts about the agent's host.
Which rule wins?
Among enforced rules, deny beats approval required, and approval required beats allow. An observing rule never changes the answer.
Read the developer section More guides
← Back to Learn